6 sock specific tcp (buffsize) useless
8 * use the exception everywhere
19 * Check the gcc version on powerpc. We disabled -floop-optimize on powerpc,
20 but versions above 3.4.0 should be ok.
28 (errors, logs ; dynars, dicts, hooks, pools; config, rrdb)
31 * Mallocators and/or memory pool so that we can cleanly kill an actor
34 * Better split casual errors from programing errors.
35 The first ones should be repported to the user, the second should kill
36 the program (or, yet better, only the msg handler)
37 * Allows the use of an error handler depending on the current module (ie,
38 the same philosophy than log4c using GSL's error functions)
41 * Hijack message from a given category to another for a while (to mask
42 initializations, and more)
43 * Allow each actor to have its own setting
44 * a init/exit mecanism for logging appender
45 * Several appenders; fix the setting stuff to change the appender
46 * more logging appenders (take those from Ralf in l2)
49 * speed up the cursors, for example using the contexts when available
52 * better formalisation of what modules are (amok deeply needs it)
53 configuration + init() + exit() + dependencies
54 * allow to load them at runtime
55 check in erlang how they upgrade them without downtime
58 * we may need a round-robin database module, and a statistical one
59 * a hook module *may* help cleaning up some parts. Not sure yet.
60 * Some of the datacontainer modules seem to overlap. Kill some of them?
67 * add the token ring as official example
68 * implement the P2P protocols that macedon does. They constitute great
72 * Spawn threads handling the communication
73 - Data sending cannot be delegated if we want to be kept informed
74 (*easily*) of errors here.
75 - Actor execution flow shouldn't be interrupted
76 - It should be allowed to access (both in read and write access)
77 any data available (ie, referenced) from the actor without
78 requesting to check for a condition before.
79 (in other word, no mutex or assimilated)
80 - I know that enforcing those rules prevent the implementation of
81 really cleaver stuff. Keeping the stuff simple for the users is more
82 important to me than allowing them to do cleaver tricks. Black magic
83 should be done *within* gras to reach a good performance level.
85 - Data receiving can be delegated (and should)
86 The first step here is a "simple" mailbox mecanism, with a fifo of
87 messages protected by semaphore.
88 The rest is rather straightforward too.
90 * use poll(2) instead of select(2) when available. (first need to check
91 the advantage of doing so ;)
93 Another idea we spoke about was to simulate this feature with a bunch of
94 threads blocked in a read(1) on each incomming socket. The latency is
95 reduced by the cost of a syscall, but the more I think about it, the
96 less I find the idea adapted to our context.
98 * timeout the send/recv too (hard to do in RL)
100 * multiplex on incoming SOAP over HTTP (once datadesc can deal with it)
102 * The module syntax/API is too complex.
103 - Everybody opens a server socket (or almost), and nobody open two of
104 them. This should be done automatically without user intervention.
105 - I'd like to offer the possibility to speak to someone, not to speak on
106 a socket. Users shouldn't care about such technical details.
107 - the idea of host_cookie in NWS seem to match my needs, but we still
108 need a proper name ;)
109 - this would allow to exchange a "socket" between peer :)
110 - the creation needs to identify the peer actor within the process
112 * when a send failed because the socket was closed on the other side,
113 try to reopen it seamlessly. Needs exceptions or another way to
114 differentiate between the several system_error.
115 * cache accepted sockets and close the old ones after a while.
116 Depends on the previous item; difficult to achieve with firewalls
119 * Implement gras_datadesc_cpy to speedup things in the simulator
120 (and allow to have several "actors" within the same unix process).
121 For now, we mimick closely the RL even in SG. It was easier to do
122 since the datadesc layer is unchanged, but it is not needed and
124 gras_datadesc_cpy needs to provide the size of the corresponding messages, so
125 that we can report it into the simulator.
126 * Add a XML wire protocol alongside to the binary one (for SOAP/HTTP)
130 * Inter-arch conversions
132 - Convert in the same buffer when size increase
133 - Exchange (on net) structures in one shoot when possible.
134 - Port to really exotic platforms (Cray is not IEEE ;)
135 * datadesc_set_cste: give the value by default when receiving.
136 - It's not transfered anymore, which is good for functions pointer.
138 - Cleanup the code (bison?)
139 - Factorize code in union/struct field adding
140 - Handle typedefs (needs love from DataDesc/)
141 - Handle unions with annotate
143 - Handle long long and long double
144 - Forbid "char", allow "signed char" and "unsigned char", or user code won't be
145 portable to ARM, at least.
146 - Handle struct/union/enum embeeded within another container
147 (needs modifications in DataDesc, too)
151 - Check struct { struct { int a } b; }
153 * gras_datadesc_import_nws?
156 * A proper RPC mecanism
157 - gras_rpctype_declare_v (name,ver, payload_request, payload_answer)
158 (or gras_msgtype_declare_rpc_v).
159 - Attaching a cb works the same way.
160 - gras_msg_rpc(peer, &request, &answer)
161 - On the wire, a byte indicate the message type:
162 - 0: one-way message (what we have for now)
163 - 1: method call (answer expected; sessionID attached)
164 - 2: successful return (usual datatype attached, with sessionID)
165 - 3: error return (payload = exception)
166 - other message types are possible (forwarding request, group
170 * Group communication
171 * Message declarations in a tree manner (such as log channels)?
173 [GRASPE] (platform expender)
174 * Tool to visualize/deploy and manage in RL
175 * pull method of source diffusion in graspe-slave
177 [Actors] (parallelism in GRAS)
178 * An actor is a user process.
179 It has a highly sequential control flow from its birth until its death.
180 The timers won't stop the current execution to branch elsewhere, they
181 will be delayed until the actor is ready to listen. Likewise, no signal
182 delivery. The goal is to KISS for users.
183 * You can fork a new actor, even on remote hosts.
184 * They are implemented as threads in RL, but this is still a distributed
185 memory *model*. If you want to share data with another actor, send it
186 using the message interface to explicit who's responsible of this data.
187 * data exchange between actors placed within the same UNIX process is
188 *implemented* by memcopy, but that's an implementation detail.
190 [Other, more general issues]
191 * watchdog in RL (ie, while (1) { fork; exec the child, wait in father })
192 * Allow [homogeneous] dico to be sent
193 * Make GRAS thread safe by mutexing what needs to be
194 * Use a xbt_set for gras_procdata_t->libdata instead of a dict
195 so that the search can be linear.
202 * finish this module (still missing the saturate part)
203 * add a version guessing the appropriate datasizes automatically
205 * provide a way to retrieve the host load as in NWS
206 * log control, management, dynamic token ring
207 * a way using SSH to ask a remote host to open a socket back on me